`

Provision versus Appropriation in Symmetric and Asymmetric Social Dilemmas*

James C. Cox,
1
Elinor Ostrom,
2
Vjollca Sadiraj,
1
and James M. Walker
3


August 28, 2012

Abstract:
Social dilemmas characterize decision environments in which individuals exclusive pursuit of
their own material self-interest can produce inefficient allocations. Social dilemmas are most
commonly studied in provision games, such as public goods games and trust games, in which the
social dilemma can be manifested in foregone opportunities to create surplus. Appropriation
games are sometimes used to study social dilemmas which can be manifested in destruction of
surplus, as is typical in common-pool resource extraction games. A central question is whether
social dilemmas are more serious for inhibiting creation of surplus or in promoting its
destruction. This question is addressed in this study with an experiment involving three pairs of
payoff-equivalent provision and appropriation games. Some game pairs are symmetric while
others involve asymmetric power relationships. We find that play of symmetric provision and
appropriation games produces comparable efficiency. In contrast, power asymmetry leads to
significantly lower efficiency in an appropriation game than in a theoretically equivalent
provision game. This outcome can be rationalized by reciprocal preference theory but not by
models of unconditional social preferences.


______________________________
1
Experimental Economics Center and Department of Economics, Georgia State University
2
Deceased founder of the Workshop in Political Theory and Policy Analysis, Indiana University
and the Center for the Study of Institutional Diversity, Arizona State University
3
Department of Economics and Workshop in Political Theory and Policy Analysis, Indiana
University


* Financial support was provided by the National Science Foundation (grant numbers SES
0849590 and SES0849551).
Provision versus Appropriation in Symmetric and Asymmetric Social Dilemmas
1. Introduction
Social dilemmas characterize settings where a divergence exists between expected
outcomes from individuals pursuing strategies based on narrow self-interest versus groups
pursuing strategies based on the interests of the group as a whole. A large literature in several
disciplines studies specific manifestations of social dilemma situations (Axelrod, 1981; Gautschi,
2000; Heibing, et al., 2011; Marshall, 2004). Two prominent areas in the economics literature are
public goods games and trust games. These are typically surplus creation games in which the
central question is whether free riding or absence of trust leads to an opportunity cost that a
potential surplus is not created nor provided for a group. For example, in the one period voluntary
contributions public good game reported by Walker and Halloran (2004), decision makers on
average failed to create 47 percent of the feasible surplus.
1
In the investment (or trust) game
reported by Berg, Dickhaut, and McCabe (1995), decision makers on average failed to create 48
percent of the feasible surplus.
2

The ultimatum game is a well-known surplus destruction game; in a typical game, the
entire surplus available to the two players is destroyed if the responder rejects a proposed split. In
the seminal ultimatum game study reported by Guth, et al. (1982), 10 percent of the feasible
surplus was destroyed by inexperienced subjects. This figure increased to 29 percent with
experienced subjects.
3
Another, well known example of a surplus destruction game is
appropriation from a common-pool resource. Walker, et al. (1990), report data for a multiple-
decision-round setting where players, on average, over-appropriated to the point of destroying the
entire available surplus from the common pool, consistent with the outcome referred to as the
tragedy of the commons.
4

An open empirical question is whether social dilemmas are more serious when related to
under-provision or over-appropriation in comparable environments. In the field, and most prior
laboratory studies, critical differences exist in the strategy and outcome spaces that make direct
comparisons between provision and appropriation social dilemmas infeasible. We address the
question by constructing three pairs of provision and appropriation games. The two games within
each pair are payoff equivalent for models of self-regarding (or economic man) preferences and
models of unconditional social preferences. As we explain, payoff equivalence implies
2

isomorphism and efficiency equivalence for these models; therefore, we often simply refer to pairs
of equivalent provision and appropriation games.
In the appropriation game in an equivalent pair, the value of the total endowment is: (a)
strictly greater than the value of the total endowment in the provision game; but (b) equal to the
maximum attainable total payoff in the provision game. The endowment in an appropriation game
is a group fund from which surplus-destroying extractions can be made by participants. The
endowments in a provision game are private funds, from which surplus-creating contributions to a
group fund can be made by participants. One experimental question is whether the theoretical
equivalence of these appropriation and provision games fails empirically and, if so, whether
efficiency (or realized economic surplus) is lower in the provision game or the appropriation game
in a theoretically equivalent pair. In field environments, institutions for provision and
appropriation exist within larger economic and political contexts that often involve asymmetries in
power. This motivates the treatments reported herein that focus on the implications of symmetric
versus asymmetric power in paired provision and appropriation games.
We examine strategies and outcomes in three pairs of games. Each pair consists of a
provision game and an appropriation game that have the same set of feasible allocations and
payoffs. The only difference between the two games within a pair is whether the agents initial
endowments are private property or common property. In contrast, pairs of games differ in their
types of asymmetry. In the baseline games all N agents move simultaneously. In contrast, in the
boss game N-1 workers simultaneously move first and the boss subsequently determines
everyones payoff after observing the workers play of the game. The king (being sovereign) has
even more power: the king not only moves last, after observing the (simultaneous) first moves of
the peasants, he can also appropriate all surplus created in the provision game or not previously
destroyed in the appropriation game. The design of the experiment crosses the (provision or
appropriation) game form treatments with the (baseline or boss or king) power treatments in a 2 X
3 design. This experimental design provides new insights into the ways in which (a) provision
versus appropriation and (b) power symmetry versus power asymmetry affect behavior in
environments characterized by social dilemmas. The two games within each pair are payoff
equivalent, isomorphic, and efficiency equivalent for unconditional preference theories. Self-
regarding (or economic man) preference theory and unconditional models of other-regarding
preferences, including the social preference theories of Fehr and Schmidt (1999), Bolton and
3

Ockenfels (2000), Charness and Rabin (2002), and Cox and Sadiraj (2007), predict that agents will
make choices that yield the same payoffs in the baseline provision and appropriation games. These
theories also predict that the boss (resp. king) provision game is equivalent to the boss (resp. king)
appropriation game. Unconditional preference theories do not predict that agents will make the
same choices in a boss or king game as they do in the baseline game. But the unconditional
preference theories do predict that the appropriation and provision games in each (baseline or boss
or king) pair of games are equivalent. Reciprocal preference theory has quite different
implications. Provision and appropriation games in either of the asymmetric power (boss or king)
pairs are not equivalent according to revealed altruism theory (Cox, Friedman, and Sadiraj, 2008).
That theory makes specific predictions about how play will differ in an asymmetric (boss or king)
provision game from the paired asymmetric appropriation game.

2. Provision and Appropriation Games with Symmetric and Asymmetric Power
A pair of games consists of a provision game and an appropriation game. The games can
be played by any number of agents N larger than two. In our baseline games with symmetric
power, all N players move simultaneously. In the asymmetric power games, N-1 players
simultaneously move first and one player moves second.

2.a. Simultaneous-Move (Baseline) Provision Game
The simultaneous-move provision game (PG) is a contributions game in which N agents
(simultaneously) choose amounts they will contribute from their endowed private funds to a group
fund that yields a surplus to be shared equally among all group members. Each agent is endowed
with e tokens in a private fund and can choose an amount
j
x from the set {0,1, 2, , } X e = to
contribute to the group fund. Contributions to the group fund create surplus; each token added to
the group fund decreases the value of the private fund of the contributor by $1 and increases the
value of the group fund by $M, where N > M > 1. From the perspective of the literature in
experimental economics, it is most natural to view the provision game as a linear voluntary
contributions mechanism, where contributions create a non-rival public good. In this case,
contributions create a symmetric positive externality to each group member. However, note that
one can also view provision as creating a common pool that is shared equally among group
4

members. Of course, the nature of the good (and the interpretation of the decision setting) would
be critical if one were to examine the effect of a change in group size. By definition, an increase in
group size would not alter the individual externality created by contributions in a pure public good
setting, while increasing group size would diminish the individual share of the group fund
allocated in a common-pool setting.
In summary, each agent is endowed with e tokens worth $1 each. Each token contributed
to the group fund yields $ . M

Let
j
x denote the contribution to the group fund by agent . j Each
of the N

agents chooses the number of tokens to contribute
j
x , 1, 2, j N = , from the feasible set
{0,1, 2, , } e . The dollar payoff to agent i

equals the amount of her endowment that is not
contributed to the group fund plus an equal (1/ N share) of M times the total amount contributed
to the group fund by all agents:
(1)

1
/
N
p
i i j
j
e x M x N t
=
= +


2.b Appropriation Game
In the simultaneous-move appropriation game (AG), N agents (simultaneously) decide how
much to extract from a group fund. The N agents are jointly endowed with E Ne = tokens in a
group fund. Each agent can choose an amount
j
z from the set {0,1, 2, , } Z e = to extract from the
group fund. Extractions from the group fund destroy surplus; each token removed from the group
fund increases the private fund of the extractor by $1 but reduces the value of the group fund by
$M where, as above, N > M > 1. Agents share equally in the remaining value of the group fund
after all extractions. Similar to the point made above regarding the provision game, it is most
natural to view the appropriation game as a common-pool resource game where, through
extraction, agents destroy surplus. Note, however, one could view an appropriation game as one
where agents appropriate resources that would have been available to provide a public good.
Taking resources prior to public good production destroys the surplus that would have been
created.
In summary, the group fund is endowed with E Ne = tokens worth $M each, for a starting
total surplus of $MNe. Each token extracted from the group fund increases the value of the private
fund of the extracting agent by $1 while reducing the value of the group fund by $M . Each of the
5

N agents chooses the number of tokens to extract
j
z , 1, 2, , j N = , from the feasible set
{0,1, 2, , } e to extract from the group fund. The dollar payoff to agent i equals the end value of
his private fund plus an equal (1/ N ) share of the remaining value of the group fund after the
extractions by all agents:
(2)
1
( ) /
N
a
i i j
j
z MNe M z N t
=
= +


2.c. Boss Provision Game (BPG) and Boss Appropriation Game (BAG)
In the BPG, N-1 agents (workers) simultaneously move first to choose the number of
tokens they will contribute to the group fund. Subsequently, the boss (agent j N = ) observes their
choices and then decides how much, if anything, to contribute. The bosss decision determines all
players final payoffs. Each of the N agents chooses the number of tokens to contribute ,
j
x
1, 2, j = N, from the same feasible set

as in the (baseline) PG game.
In the BAG, N-1 agents simultaneously move first to choose the number of tokens they
will extract from the group fund. Subsequently, the boss observes their choices and then decides
how much to extract, which determines all players final payoffs. Each of the N agents chooses the
amount to extract
j
z , 1, 2, , j N =

from the same feasible set

as in the (baseline) AG game.
2.d. King Provision Game (KPG) and King Appropriation Game (KAG)
In KPG, N-1 agents (peasants) simultaneously move first. Subsequently, the king (agent
j =N) observes their choices and then decides how much to contribute or how much to
appropriate from the other agents contributions. Each of the first movers chooses the number of
tokens to contribute
j
x , 1, 2, , 1 j N =

from the same feasible set

as in the PG and BPG games.
The king can choose to contribute any non-negative number of tokens up to his endowment e to
the group fund. Alternatively, the king can choose to take (in integer amounts) any part of the
tokens contributed by the N-1 peasants if it is strictly positive. Define
1
1
N
N j
j
X x

=
=

. The king can

choose an amount
N
x (to take or contribute) from the feasible set
{ , 1, 2, , }
PG N N N
K X X X e

= + + .
6

In KAG, N-1 agents simultaneously move first. Subsequently, the king observes their
choices and then decides how many of the remaining tokens (if any) to extract. Each of the N-1
first movers chooses an amount to extract
j
z , 1, 2, , 1 j N = from the same feasible set as in the
AG and BAG games. The king chooses an amount
N
z

to extract from the feasible set of integers
1
1
{0,1, 2, , }
N
AG j
j
E z

=
K =

.

3. Theory of Provision and Appropriation Games
We use I

to denote game (N,

,{ }
i i i
G )where N denotes the set of players,
i
G denotes
the (opportunity) strategy set of player i , i eN, and
i
represents the preferences of player i
over G = .
i i
G

As conventionally done, vectors will be represented by letters written in bold font.
5


3.1 Definitions of Equivalence
We focus on three ways in which the provision and appropriation games could be
equivalent: they could be payoff equivalent, isomorphic, or efficiency equivalent. Terms such as
efficiency and isomorphism are frequently used in the literature, although with varying
specific interpretations. We begin by providing definitions of terms that we will use in the
subsequent theoretical and empirical analysis; these terms include payoff equivalence,
isomorphism and efficiency equivalence.
We say that two games, I and u are payoff equivalent if for each player i eN there
exists some bijection,
i
P of strategy set
i
G in game I to strategy set
i
F in game u such that the
vector of such bijections preserves all players payoffs. Thus if we let P(g) denote ( ( ) :
j j
P g j eN)
one has the following formal definition of payoff equivalence:

Definition 1: Payoff Equivalence. Two games I and u are payoff equivalent if for all i eN
there exist bijections,
i
P :
i i
G F

such that for all g G e ,
i
t (g) =
i
t (P(g) ) .

7

It will help to illustrate this definition using the N-player games in this paper. As above, let
e be the private fund endowment of each player in the provision game and E

be the endowment
of the group fund in the appropriation game; by design . = E Ne Also, let
j
x denote the
contribution by agent j to the group fund in the provision game and
j
z denote that agents
extraction from the group fund in the appropriation game. It can be easily verified that mappings,
P
i
of x
i
to z
i
such that (*)
i
e x =
i
z for , e i N that is ( )
i i i
P x e x = are bijections that satisfy the
payoff equalities in Definition 1. In words, the bijection for agent i specifies that the amount
i
e x left in the private fund by agent i in the provision game equals the amount
i
z appropriated
from the group fund by that agent in the appropriation game. It is straightforward to show (see
statements (1) and (2)) that, if (*)
i
e x =
i
z for all i eN, then each player receives the same payoff
in the provision game as she does in the appropriation game (although different players may
receive different payoffs).
Next, we give a formal definition of isomorphic games; these are games for which there
exist bijections that preserve preferences of each individual player:

Definition 2: Isomorphism. Two games I and u are isomorphic if for all i eN there exist
bijections,
i
I :
i i
G F such that for all g, h G e , if g
i
I
h then I (g)
i
u
I (h).

For games included in this paper, bijections ( )
i i i
P x e x = , also work as illustrations of Definition
2.
6
It is straightforward to show (see appendix 1) that isomorphism preserves pure strategy Nash
equilibria, in the sense that g is a pure strategy Nash equilibrium in one game if and only if P(g) is
a pure strategy Nash equilibrium in the other game.
Comparisons across pairs of (baseline, boss, or kings) games raise the question of
efficiency. Unlike the previous two definitions that are not related to equilibria, the definition of
efficiency equivalence applies to equilibrium outcomes. We say that two games are efficiency
equivalent if the most efficient equilibria preserve the total payoff. Let the most efficient
equilibrium, h* in a game be the Nash equilibrium with the largest total payoff to all players, that
is
8

1
N
i
i
t
=

(h*)
1
N
i
i
t
=
>

(g), g ( ) eO I
where ( ) O I is the set of all pure strategy Nash equilibria of I .

Definition 3: Efficiency Equivalence. Games I and u are efficiency equivalent if, for the most
efficient equilibria g* and h* in I and u,
N
t
e
i
i
( g*) =
N
t
e
i
i
( h*).

3.2 Equivalence of the Simultaneous-Move Provision and Appropriation Games

Statements (1) and (2) imply that if
j j
z e x = for j eN then agent i s payoff in the
provision game (
PG
i
t ) equals her payoff in the appropriation game (
AG
i
t ), that is
PG
i
t (x

)

=
AG
i
t (e - x), and this is true for i eN. Since both, (fixed) social preferences and homo economicus
preferences, are defined on money payoff space, g
PG
i
h in the provision game if and only if e - g
AG
i
e h in the appropriation game, as stated in Proposition 1, part b. It is expected from the
discussion below Definition 2 then isomorphism and payoff equivalence imply efficiency
equivalence.

Proposition 1. The simultaneous-move provision and appropriation games are:
a. payoff equivalent;
b. isomorphic; and
c. efficiency equivalent.
Proof: See appendix 1.

3.3 Equivalence and Non-equivalence of the Sequential-move Games
Theoretical properties of sequential-move (boss and king) provision and appropriation
games depend on the distinction between fixed preferences and reciprocal preferences. The most
familiar type of fixed preferences are those for the homo economicus model in which an agents
preferences vary with her own material payoffs but are invariant with the material payoffs of
9

others. Other fixed preferences models relax the material selfishness property. Examples are given
by models of social preferences such as Fehr and Schmidt (1999), Bolton and Ockenfels (2000),
Charness and Rabin (2002), and Cox and Sadiraj (2007). In these homo economicus and social
preferences models, the preferences of an agent are a fixed characteristic of the agent that is
independent of the actions of other agents. For such preferences the boss (resp. king) provision
game is equivalent to the boss (resp. king) appropriation game, as stated in the following
proposition.

Proposition 2. For fixed (homo economicus and social) preferences, the sequential-move boss
(resp. king) provision and boss (resp. king) appropriation games are:
d. payoff equivalent;
e. isomorphic; and
f. efficiency equivalent.
Proof: See appendix 1.
We have compared the provision game with the appropriation game in each pair of
baseline, boss, and king games. Next, we compare the boss provision (resp. appropriation) game
with the king provision (resp. appropriation) game. Recall that, for a given (provision or
appropriation) game form the only difference between the (sequential) boss game and the
(sequential) king game is that at any given information set the bosss opportunity set is a strict
subset of the kings opportunity set. Hence, the cardinality of the kings set of strategies is strictly
larger than the cardinality of the bosss set of strategies. This implies that boss and king (provision
or appropriation) games are neither isomorphic nor payoff-equivalent because bijections over
finite sets preserve cardinality of the set. The remaining question is how they rank with respect to
efficiency.
This question can be answered for agents whose preferences satisfy a type of translation
invariance in the space of money payoffs. We say that a translation is an inequality-reducing one
if it decreases the money payoff of the richest players and increases the money payoffs of others. It
is costless if the total amount by which the richest individuals payoffs are reduced is not greater
than the sum of increased payoffs of others. The translation is order-preserving if identity of the
richest individuals is preserved by the translation. Let W() denote the set of richest individuals in
10

an allocation and let W
C
() denote the complement of W() in the set of players N that contains
N agents. We use notation in which R is the number of richest players (the cardinality of set W())
and, hence, the number of other players is N-R (the cardinality of set W
C
()). A formal definition
follows.

Definition 4: (utilitarian translation) For any given payoff vector , a translation v =(a,b) such
that for all ( )
C
i W e t and all ( ) j W e t
(inequality-reducing) 0 , ,
i j
v a b v i j = > > =
(efficiency-increasing) + > ( ) 0 a N R bR
( rank-preserving)
( )
max
C
i j
i W
a b
t
t t
e
+ s +
is called -utilitarian translation. The set of such translations is denoted by ( ). T t
We say that agents prefer utilitarian translations (PUT) if for all translations
( ) ( ') T T e v t t , for all , '
N
R e t t , and for all ( ) ( ') i W W e t t
(a)
i
+ t t v
(b)
*
i
t t implies
*
i
+ + v v t t .
This property of preferences means that for any individual who is not among the richest: (a) an
allocation that results from a translation e v ( ) T t is preferred to the original allocation; and (b) if
allocation of payoffs t is preferred to allocation ' t then a translation e v ( ) T t

of t

is preferred
to the translation e v ' ( ) T t

of ' t . Preferences that satisfy this property include the inequality-
averse preferences in Fehr and Schmidt (1999) and the quasi-maximin preferences in Charness
and Rabin (2002).
Proposition 3 below states that king games cant be more efficient than boss games for
preferences with property PUT.

Proposition 3. For sequential-move finite (provision or appropriation) games one has:
a. boss and king games are neither payoff equivalent nor isomorphic; and
b. for preferences with property (IUT), boss games are (weakly) more efficient than king
games.
11


Proof: See appendix 1.
We have compared (provision vs. appropriation) game forms and (boss vs. king) power
treatments with fixed (homo economicus or social) preferences models. We now consider the
implications of reciprocity. Fixed preferences are fundamentally different from reciprocal
preferences in which an agents other-regarding preferences can be dependent on the prior actions
of other agents. Cox, Friedman, and Sadiraj (2008) presents a model of reciprocity based on two
partial orderings, of opportunity sets and of preferences, and two axioms that link the partial
orderings. The theory focuses on willingness to pay (WTP) amounts of ones own material payoff
(my income, m) to change anothers material payoff (your income, y). A partial ordering of
preferences is defined for WTP, as follows. If WTP
A
(m,y) > WTP
B
(m,y) for all ( , ) m y in some
specified domain then preference ordering A is more altruistic than preference ordering B on that
domain. In that case, one writes A MAT B.

Two different preference orderings, A and B can
represent the preferences of two different agents or the preferences of the same agent in two
different situations (that differ, for example, with respect to reciprocity-eliciting actions of
someone else).
The other partial ordering is of opportunity sets. In sequential games a first movers
(your) action affects a second movers (my) opportunities. If a first movers action increases
the second movers maximum possible payoff then the second mover will regard him as generous
unless it is the case that the first mover increases his own maximum possible payoff even more, in
which case the real intention of the first mover may be just to benefit himself. For any given
opportunity sets G and , F let
G
m
-
and
F
m
-
, respectively, denote a second movers (my)
maximum possible payoffs in G and . F Let
G
y
-
and
F
y
-
denote a first movers (your)
maximum payoffs in the two sets. Opportunity set G is more generous than opportunity set F
for me (the second mover) if: (a) 0
G F
m m
- -
> and (b)
G F
m m
- -
> .
G F
y y
- -
In that case, one writes
G MGT . F Part (a) states that a first movers choice of G rather than F (weakly) increases the
second movers maximum possible payoff. Part (b) states that the first movers choice of G rather
than F did not increase his own maximum payoff more than it did the second movers maximum
payoff, thus clearly revealing generosity.
12

The essential property of reciprocal preferences is that a second movers WTP can depend
on a first movers prior actions, as represented by Axiom R in Cox, Friedman and Sadiraj (2008).
Axiom R states that if you choose my opportunity set , G when you could have chosen set , F

and
it is the case that G MGT , F then my preferences will become more altruistic towards you. More
formally, let
G
A denote my preferences when you choose (my) opportunity set G and
F
A denote
my preferences when you choose set . F Axiom R states: if G MGT F then
G
A MAT .
F
A
Reciprocal preferences can also depend on the distinction between acts of commission and
acts of omission or no opportunity to act as in Axiom S of Cox, Friedman, and Sadiraj (2008, p.
41). An informal description of Axiom S is that it says that the effect of Axiom R is stronger when
a generous act (of commission) overturns the status quo than when an otherwise same act (of
omission) merely upholds the status quo and yet stronger still than when there was no opportunity
to act.
An extension to 2 N >

players of the model in Cox, Friedman, and Sadiraj (2008) shows
that play by second movers is predicted to be different in sequential provision games than in
sequential appropriation games. It is straightforward to use the (above) definition of the MGT
partial ordering of opportunity sets to show the following: (a) the second movers opportunity set
in the king (resp. boss) appropriation game is the most generous possible if the first movers do not
change the common-pool endowment (i.e. they appropriate nothing for their private funds); (b)
each additional token that any first mover appropriates in the king (resp. boss) appropriation game
makes the second movers opportunity set incrementally less generous than it was, which
according to Axioms R and S makes the second mover less altruistic than he was; (c) the second
movers opportunity set in the king (resp. boss) provision game is the least generous possible if the
first movers do not change their private endowments (i.e. provide nothing to the group fund); and
(d) each additional dollar that any first mover provides to the group fund makes the second
movers opportunity set incrementally more generous than it was, which according to Axioms R
and S makes the second mover more altruistic than she was.
Let e g
-N
be the vector of amounts left in their private funds (i.e. not provided to the group
fund) by first movers in a sequential provision game and let g
-N
be the vector of appropriations
from the group fund in a sequential appropriation game. In that case, Axioms R and S imply that
the second movers preferences will not be the same in the king (resp. boss) appropriation game as
13

in the king (resp. boss) provision game because the (same) second mover opportunity set resulted
from an ungenerous change from the endowed opportunity set in the sequential (boss or king)
appropriation game and a generous change from the endowed opportunity set in the sequential
(boss or king) provision game. This intuition is formalized in the following proposition about the
second movers best response in the sequential provision game (
P D G
br ) and in the sequential
appropriation game (
A D G
br ), where D = K (for king) or B (for boss).

Proposition 4. Assume reciprocal preferences characterized by Axioms R and Axiom S. At any
given information set I (g
-N
) determined by first movers choices g
-N
, the second movers optimal
choice is more efficient in the king (resp. boss) provision game, DPG than in the corresponding
king (resp. boss) appropriation game, DAG. That is, for all g
-N
i
i N
G
<
e ,
DAG
e br (eg
-N
) s
DPG
br
(g
-N
), where D = K or B.

Proof: See appendix 1.

4. Experiment Results
Experiment sessions were conducted at both Georgia State University and Indiana
University.
7
In each session, subjects were recruited from subject data bases that included
undergraduates from a wide range of disciplines. Via the computer, the subjects were privately and
anonymously assigned to four-person groups and remained in these groups throughout the session.
No subject could identify which of the others in the room was assigned to their group. Since no
information passed across groups, each session involved numerous independent groups. At the
beginning of each session, subjects privately read a set of instructions that explained the decision
setting. In addition, an experimenter reviewed the instructions publicly. The games described
above were operationalized in a one-shot decision setting with a double-blind payoff protocol. The
game settings and incentives were induced in the following manner.
In the PG treatment, each individual is endowed with 10 tokens worth $1 each in what was
referred to in the experiments as his/her Individual Fund. The decision task is whether to move
tokens to a Group Fund. Any tokens moved to the Group Fund are tripled in value. Individual
earnings equal the end value of the Individual Fund plus one-fourth of the end value of the Group
14

Fund. Second movers in the BPG and KPG treatments are allowed choices as described in section
2.
In the AG treatment, each group is endowed with 40 tokens worth $3 each in their Group
Fund. The choice of each individual is whether to move tokens to his/her own Individual Fund.
Any tokens moved from the Group Fund reduce the value of the Group Fund by $3, and increase
the value of the Individual Fund of the decision maker by $1. Individual earnings equal the end
value of the Individual Fund plus one-fourth of the end value of the Group Fund. The second
movers in the BAG and KAG treatments are allowed choices as described in section 2.
The Nash equilibrium for the special case of homo economicus preferences would call for
each subject to make a zero contribution to the Group Fund in all of the provision treatments. In
the appropriation treatments, the equilibrium entails each subject extracting 10 tokens from the
Group Fund. In contrast, the group optimum occurs when all tokens are contributed to the Group
Fund in a provision game and when no tokens are extracted from the Group Fund in an
appropriation game. Unconditional social preferences models predict that the number of tokens
contributed in a provision game is the same as the number of tokens not extracted in the paired
appropriation game.
Data are reported for the number of individual subjects (and four person groups) listed in
Table 1. We doubled the initial sample size for the KPG and KAG treatments after observing a
striking difference (reported below) between these treatments, in order to ensure that this result
was not due to a small sample bias.
The summary presentation of results focuses on three primary behavioral characteristics of
the experiments: (1) efficiency or variation in payoffs across the six treatment conditions; (2)
choices by first movers in all treatments; and (3) choices by second movers in the four sequential
treatment conditions.
4.a. Realized Surplus
The most fundamental issue related to the alternative treatment conditions is the impact of
game form on the ability of group members to generate surplus in the three provision games and
not to destroy surplus in the three appropriation games. Using each four-member group as the unit
of observation, note that both the minimum possible group payoff ($40) and the maximum
15

possible group payoff ($120) are constant across all six treatments. Figure 1 displays average
group earnings in the six treatment conditions.
Result 1: Average group earnings across the two baseline conditions (PG and AG) are
very similar. Earnings are well above the minimum predicted by Nash equilibrium for the
special case of homo economicus preferences (which is $40).

Result 1 is consistent with Proposition 1. Also, the data for the baseline PG treatment are
consistent with findings from a large number of linear public goods experiments: the complete
free riding prediction from the homo economicus preferences model fails empirically. Also, the
data for the baseline AG treatment are inconsistent with a strong form tragedy of the commons
prediction, from the self-regarding preferences model, that all available surplus will be destroyed.

Result 2: Average earnings are lower in the asymmetric power BPG and BAG treatments
than in the symmetric power PG and AG treatments, and are even lower in the asymmetric
power KPG and KAG treatments.

The first part of result 2 is inconsistent with theoretical implications for fixed preferences models,
as stated in Proposition 2. Power asymmetries decrease efficiency (or realized surplus) in both
provision and appropriation settings. Low efficiency is especially a feature of the king treatment
for the appropriation setting, which is consistent with Proposition 3. Treatment KAG comes
closest to manifesting a strong form tragedy of the commons.

Result 3: Pooling across decision groups (n=70), least squares analysis of total group
allocations to the Group Fund leads to the following results related to selective tests
of equality. Earnings differences between treatments in provision settings are statistically
significant for PG vs. BPG and for PG vs. KPG. Earnings differences between treatments
in appropriation settings are significant for AG vs. KAG and for BAG vs. KAG. Earnings
are significantly lower for KAG than for KPG.
8

4.b. Type X Decisions
16

For comparison purposes, the decisions of Type X subjects (all subjects in the
simultaneous PG and AG games, and those randomly assigned to be first movers in the sequential
games) are presented as the dollar amounts allocated to the Group Fund in the provision games or
dollar amounts left in the Group Fund in appropriation games. In the notation of section 2, the bar
graph shows the average value across Type X subjects of 3
j
x in provision games and 30
j
z in
appropriation games.
Result 4: Pooling across Type X decision makers (n=227), least squares analysis of Type
X token allocations to the Group Fund (tokens left in the Group Fund) leads to the
following results related to selective tests of equality. Group Fund differences between
treatments in provision settings are statistically significant for PG vs. KPG. Group Fund
differences between treatments in appropriation settings are significant for AG vs. KAG.
9


4.c. Type Y Decisions
Figure 3 displays the decisions of the second movers (Type Y) for the four treatments with
sequential decision making. For the boss treatments, decisions are represented as average dollar
amounts contributed to the Group Fund (BPG setting) or left in the Group Fund (BAG setting).
For the KPG treatment, the bar graph shows the average value across Type Y subjects of
4
3x ,
where
4
x is a non-negative number of tokens up to 10 or a negative (tokens withdrawn) number
up to the maximum number of tokens contributed by the three Type X players in her group. For
the KAG treatment, the bar graph show the average across Type Y subjects of the amount (30-
3z
4
), where
4
z is the amount withdrawn from the Group Fund, chosen from the feasible set
3
1
{0, 3, 6, ,120 }
CPR j
j
z
=
K =

.
10

We next report an analysis of Type Y token allocations using treatment dummy variables.
Pooling across Type X decision makers (n=53), a tobit regression of Type Y decisions was
conducted controlling for the total Type X token allocation (XSUM) to the Group Fund (or total
tokens left in the Group Fund) and treatment and location dummy variables

17

Result 5: Only one coefficient estimate is statistically significant, the negative coefficient
for the dummy variable for the KAG treatment. The coefficient for the KPG treatment is
negative but insignificant. After controlling for treatment effects, the amount first movers
place in the Group Fund in provision settings or leave in the Group Fund in appropriation
settings is not a significant determinant of second mover choice.
11

The insignificance of the coefficient for first movers contributions is weakly consistent with
homo reciprocans theory but also consistent with fixed (non-reciprocal) social preferences models.
In contrast, the significance of the coefficient for the KAG treatment is consistent with reciprocal
preferences, as stated in Proposition 4, but inconsistent with the implications of fixed preferences
as stated in Proposition 2.
12

4.d. Type X and Type Y Earnings
Table 2 displays average earnings for each treatment condition. Not surprisingly, the
actions by second movers in the asymmetric power games are of particular importance in
determining final earnings. Consistent with the discussion above, the decisions by second movers
in the kings treatment, especially KAG, creates a large discrepancy in earnings between first and
second movers.
In summary, the analysis of data from these experiments suggests that the opportunity for
second movers to exploit cooperative decisions by first movers: (a) significantly reduces first
movers level of cooperation (and resulting efficiency); and (b) leads second movers to exploit
this opportunity, in particular in the kings setting.

5. Concluding Remarks
In this paper we report differences in behavioral outcomes in two types of social dilemmas:
provision games and appropriation games. We examine (baseline) symmetric games where all
players act at the same time without knowing what others contribute (to a public good) or
appropriate (from a common pool) and two types of asymmetric (boss and king) games. In the
boss game, three players act first and, subsequently, with knowledge of their decisions the fourth
player decides all players final payoffs by choosing how much to contribute or appropriate. In the
king treatment, three players act first, and with knowledge of their decisions, the fourth player
18

decides how much to contribute or take when given the capability of taking everything. While
participants do contribute (or refrain from taking) more than predicted in the special-case homo
economicus interpretation of game theory in the symmetric condition, average payoffs fall
significantly when one player has asymmetric power. The presence of a fourth actor (a king)
who can take resources contributed (to a public good) by others or take resources left (in a
common pool) by others has a strong adverse effect on the total payoff in a game. In particular,
with a king present in the surplus appropriation setting, we observe outcomes that closely
approximate the tragedy of the commons.
The experiment provides tests of the different implications of fixed (homo
economicus or social) preferences models and of a model of reciprocal preferences for behavior in
provision (public good) and appropriation (common pool) social dilemmas. The provision and
appropriation games in each pair of (baseline, boss, or king) games are (by design) payoff
equivalent, isomorphic, and efficiency equivalent for fixed preferences models. Therefore, such
models predict the same outcomes for the provision and appropriation games in each pair of
games (although different outcomes across pairs). In contrast, the reciprocity model in revealed
altruism theory predicts specific differences in outcomes between the provision and appropriation
games in each pair of asymmetric power (boss or king) games. Data from the boss and king
treatments are generally consistent with implications of the reciprocity model but inconsistent with
implications of the fixed preferences models
19

References
Axelrod, Robert (1981). The Emergence of Cooperation Among Egoists. American Political
Science Review 75(2): 306-318.
Berg, Joyce, John Dickhaut, and Kevin McCabe. 1995. Trust, Reciprocity, and Social History,
Games and Economic Behavior 10: 122-142.
Bolton, G. E. and A. Ockenfels. 2000. ERC: A Theory of Equity, Reciprocity, and Competition.
American Economic Review 90(1): 16693.
Charness, Gary and Matthew Rabin. 2002. Understanding Social Preferences with Simple Tests.
Quarterly Journal of Economics 117: 81769.
Cox, James C., Daniel Friedman, and Vjollca Sadiraj. 2008. Revealed Altruism. Econometrica
76(1): 3169.
Cox, James C. and Vjollca Sadiraj. 2007. On Modeling Voluntary Contributions to Public
Goods. Public Finance Review 35(2): 31132.
Fehr, Ernst and Klaus Schmidt. 1999. A Theory of Fairness, Competition, and Cooperation.
Quarterly Journal of Economics 114(3): 81768.
Guth, Werner, Rolf Schmittberger, and Bernd Schwarze, 1982. An Experimental Analysis of
Ultimatum Bargaining, Journal of Economic Behavior and Organization 3: 367-388.
Gautschi, Thomas (2000). History Effects in Social Dilemma Situations. Rationality and Society
12(2): 131-162.
Heibing, Dirk, Wenjian Yu, Heiko, Rauhut (2011). Self-Organization and Emergence in Social
Systems: Modeling the Coevolution of Social Environments and Cooperative Behavior.
The Journal of Mathematical Sociology 35(1): 177-208.

Marshall, Graham R. (2004). Farmers Cooperating in the Commons? A Study of Collective
Action in Salinity Management. Ecological Economics 51: 271-286.

Walker, James and Matthew Halloran, 2004. Rewards and Sanctions and the Provision of Public
Goods in One-Shot Settings, Experimental Economics 7: 235-247.

Walker, James, Roy Gardner, and Elinor Ostrom, 1990. Rent Dissipation in Limited Access
Common Pool Resource Environments: Experimental Evidence. Journal of
Environmental Economics and Management 19: 203-211.

20

Table 1. Number of Individual Subject (and Group) Observations by Treatment
Symmetric
Power
Boss
Asymmetric Power
King
Asymmetric Power

Provision Games
32
(8 Groups)
28
(7 Groups)
76
(19 Groups)

Appropriation Games
36
(9 Groups)
32
(8 Groups)
76
(19 Groups)


Table 2. Experimental Earnings by Subject Type and Treatment Condition
Symmetric
Power
Type X
Boss
Asymmetric Power
Type X Type Y
King
Asymmetric Power
Type X Type Y

Provision Games $24.19

$19.53

$21.43

$16.43

$21.92

Appropriation Games

$22.39

$20.84

$21.31

$11.04

$22.10







21




Figure 1. Total Earnings for Groups by Treatment
PG BPG KPG AG BAG KAG
Maximum Possible Payoffs
Minimum Possible
Payoffs
$120
$40
$100
$80
$60
$20
$96.76
$80.00
$71.26
$89.56
$83.50
$54.74












22




Figure 2. Average Individual Type X Decisions Represented as $ in Group Fund

PG BPG KPG AG BAG KAG
Maximum Possible Contribution
12
$30
24
18
6
$21.27
$16.44
$15.84
$18.57
$16.74
$13.41


23

Figure 3. Average Individual Type Y Decisions Represented as $ in Group Fund

BPG KPG BAG KAG
-15
$30
15
0
-30
$10.62
$15.00
$-18.16
$-0.63
Maximum Group Fund Contribution
Average
Minimum
Contribution
Equals -$40.23
Average
Minimum
Contribution
Equals -$47.52
Minimum
Contribution
Equals $0.00
Minimum
Contribution
Equals $0.00

























24

Endnotes

1
Calculation based on overall baseline VCM, Table 2, page 240.

2
Calculation based on average amount sent by first movers, Figure 2, page 130.

3
Calculation based on percent of rejected offers, Tables 4 and 5, page375.

4
Calculation based on percentage of rents earned in the high endowment setting, Table II, page 208.

5
We will use N also as a notation for the cardinality of the set of players.

6
It is easy to show that isomorphic games are not necessarily payoff-equivalent. One example would be splitting $10
and splitting $20 on a continuous space; or in a discrete space, splitting $10 in a one-cent unit of divisibility and
splitting $20 in a two-cent unit of divisibility. The splitting $10 and $20 games are clearly isomorphic but not payoff-
equivalent.

7
Complete subject instructions for the experiment are available at http://excen.gsu.edu/prov_approp_games/ .

8
All linear model analyses are conducted with robust standard errors: PG = BPG, p = .05; PG = KPG, p = .00 ; BPG =
KPG, p= .11; AG = BAG, p = .71; AG = KAG, p = .00; BAG = KAG, p= .00; PG = AG, p = .07; BPG = BAG, p =
.66; KPG = KAG, p = .04; Lab Location (GSU versus IU), p=.14.

9
VCM = BVCM, p = .16; VCM = KVCM, p = .01; BVCM = KVCM, p= .52 ; CPR = BCPR, p = .78; CPR = KCPR,
p = .02; BCPR = KCPR, p= .13; VCM = CPR, p = .20; BVCM = BCPR, p = .92; KVCM = KCPR, p = .28; Lab
Location (GSU versus IU), p=.19.
10
The maximum amount that can be withdrawn by a Type Y subject in either the KPG or KAG treatment depends on
the decisions by Type X subjects in his/her decision making group in the relevant treatment session. The Type X data
reported in Figure 2 show the average constraint faced by the Type Y subjects in Figure 3.

11
With PG as the baseline setting for the analysis, we find the following parameter estimates and level of significance:
DUMBCPR: 2.71 (p=.71); DUMKCPR: -15.47 (p=.03); DUMKVCM: -7.25 (p=.29); DUMIU: 5.99 (p=.21); XSUM:
-0.29 (p=.34); CONSTANT: 10.99 (p=.13).

12
By design, in KAG, Type Y subjects make only one decision, how many tokens to remove from the Group Fund. In
KPG, Type Y subjects chose whether to add tokens to the Group Fund from their Individual Fund or remove tokens
from the Group Fund. To examine the robustness of our initial findings, we designed an alternative setting, KPG2. In
KPG2, groups began with 10 tokens in the Group Fund rather than zero. Type Y subjects began with zero tokens in
their Individual Fund, and Type Y subjects made only one decision, how many tokens to remove from the Group
Fund. Data from 17 groups was collected. Comparing the data from KPG and KPG2, no significant difference was
found in regard to Type X decisions (two sided p-value is 0.102) or Type Y decisions (two sided p-value is 0.662).
1 Appendix 1: Proofs of Propositions
Lemma 1 Isomorphism preserves pure Nash equilibria.
Proof. Suppose that games ( and T are isomorphic: ( , T. Then, there
exists bijective functions, 1
i
: G
i
1
i
that preserve preference ordering. Let
A(() and A(T) be the sets of pure strategy Nash equilibria of games ( and
T. We need to show that g A(() if and only if f = I(g) A(T). Since we
are dealing with bijections it suces to show in one direction. Let i N and
/ G
i
be given. We show that f _
i
(/, f
i
). By surjectivity of 1
j
there exists
g
0
G for which I(g
0
) = (/, f
i
) and by injectivity of 1
j
, g
0
j
= q
j
for all , ,= i.
The last statement and g A(() imply that g _
i
g
0
and by the denition of
isomorphism I(g) _
i
I(g
0
), hence f _
i
(/, f
i
).
Lemma 2 For unconditional preferences, payo equivalence implies isomor-
phism and eciency equivalence.
Proof. Suppose that games ( and T are payo equivalent. Then for all i
there exists bijections 1
i
: G
i
1
i
such that P = (1
1
, ..., 1
N
) preserves the
payos.
We need to show that for all g, h G, g _
i
h i P(g) _
i
P(h). It follows
from payo equivalence, that for f g, h, payo distribution of P(f ) is iden-
tical to the payo distribution of f . Since unconditional preferences are de-
ned solely over payo distributions, for all i N, g _
G
i
h and
i
(f )
i2N
=

i
(1(f ))
i2N
, f g, h imply that P(g) _
i
P(h).
Next, let g

be one of the most ecient Nash equilibria in (. Then, by

Lemma 1, P(g

) is a Nash equilibrium in T and by payo equivalence,

P
i

i
(g

) =
P
i

i
(P(g

)). Thus, there exists a Nash equilibrium in T with same total payo
as g

, hence ( cannot be more ecient than T. To show that T cannot be more

ecient than ( take 1
i
= 1
1
i
, i and apply the same arguments.
Proof of Proposition 1. Let T and / denote a provision game and its
corresponding appropriation game. For both games the strategy sets, 1 and
for each player i satisfy 1 = [0, c] = . For all i , consider function T
i
:
1 such that T
i
(r) = c r.
To show payo-equivalence consider T :
i
1
i
such that T(g) =
(T
i
(q
i
)
i2N
). This function is well dened because (T(q
i
))
i

N
. We need to
verify that transformation T is a bijection that preserves payos. To show
injectivity, take any g, h
i
1 such that g ,= h. This means that there exists
some i such that q
i
,= /
i
, hence T(q
i
) = cq
i
,= c/
i
= T(/
i
), so T(g) ,= T(h).
To show surjectivity for any f
i
take g = (c )
1
, c )
2
, ..., c )
N
) and
verify that T(g) = f . Finally, T preserves payos for each player i since by
1
construction

i
(T(g)) = T
i
(q
i
) +,(1
X
j
T
j
(q
j
))
= c q
i
+,
X
j
(c (c q
j
)) = c q
i
+,
X
j
q
j
=
i
(g)
It follows from payo equivalence and Lemma 2 that these two games are
isomorphic and neither is more ecient.
The proof for the sequential versions of the games is similar.
Proof of Proposition 2. Consider a provision game. We show that sequential
provision game, T
s
is (weakly) more ecient than the simultaneous provision
game, T, which is the same ecient as the corresponding appropriation game;
that is
T
s
_ T - /
Let h

be a most ecient (pure) Nash equilibrium of T. We show that there

exists a Nash equilibrium of the sequential game that is the same ecient (in the
sense that the total payo is the same) as h

. Without any loss of generality let

1 -player be player and consider the following strategy, j: contribute /

N
at
all information sets. Then, the strategy prole (h

N
, j) is a Nash equilibrium in
the sequential game and it is the same ecient as, h

. Hence, the simultaneous

game cannot be more ecient than the sequential provision game.
Similarly,
/
s
_ / - T
Thus, for xed preferences, sequential games are (weakly) more ecient than
simultaneous games.
Proof of Proposition 3. Part (1) follows from strategy sets of Y-player being
discrete and having dierent cardinality in the king and boss game whereas
bijections preserve cardinality. We show part 2 for the provision games; apply
isomorphism (or similar arguments to the ones below) to show that the results
hold for the appropriation games as well.
Let a most ecient Nash equilibrium in the king game be
K

= (x

, 1

(.))
where 1

(.) is the strategy function of the Y-player. If we let A(x) denote

P
iN
r
i
then at any information set J(x) in the king game
1

(x) [A(x), c]
Let j

denote Y-players contribution 1

(x

) at the information set J(x

)
(i.e. on the equilibrium path) and consider prole
k

= (x

, i(x))
2
with i(.) such that \x [0, c]
N1
,
i(x) = j

, if j

_ A(x)
= 1

(x), otherwise
Prole k

diers from prole K

with respect to Y-choices only in information

sets J(x) in which j

is a feasible choice; in such sets Y-choices are all j

. By
construction, k

is eciency-equivalent to K

and a Nash equilibrium in the

king game.
Next, consider strategy prole b = (x

, c(x)) in the boss game that diers

from prole k

only with respect to Y-strategy which is dened as follows c(x) =

maxi(x), 0, \x [0, c]
N1
. The total payo of b is not smaller than the total
payo of k

since by construction,
X

i
(b) _
X

i
(k

). To complete the
proof we need to show that for preferences with property (xxxx), b is a Nash
equilibrium in the boss game.
We distinguish between two possible cases: j

_ 0 and j

< 0. In the former

case, i(x) = j

= c(x), so b = k

is a Nash equilibrium in the boss game as

well, which completes the proof.
Suppose that j

< 0. To show that Y-player cannot do strictly better by

deviating it suces to verify that at information set J(x

), c(x

) = 0 is opti-
mal. It follows from k

being a Nash equilibrium in the king game that (for

well-behaved preferences of Y-player, i.e. n
N
concave on player Ys con-
tribution j) the 1 -players utility must be decreasing on (j

, c]. So, if j

< 0
then 0(= c(x

)) is the best choice 1 -player can make at J(x

) when the feasi-

ble choice set is [0, c]. Next, we show that player 1s choice r

1
is optimal. Let
(q, x
1
, i(q, x
1
)) denote the payo vector generated by the vector of contri-
butions (q, x
1
, 1(q, x
1
)). Formally one has
(q, x
1
, i(q, x
1
)) = (c + (q) q, c + (q) x
1
, c + (q) i(q, x
1
)) (1)
where c = c+ ,
X
j=2::N1
r
j
and (q) = ,(q +i(q, x
1
)). Strategy prole k

is a
Nash equilibrium in the king game, so
(x

, j

) _
1
(q, x

1
, i(q, x

1
)), \q [0, c] (2)
We need to show that
(x

, 0) = (x

, c(x

)) _
1
(q, x

1
, c(q, x

1
)), \q [0, c] (3)
since by construction, i(x

) = j

(< 0) and c(x

) = 0. Let q [0, c] be given.

There can be two cases only: (a) i(q, x

1
) = j

and (b) i(q, x

1
) j

.
Case (a): i(q, x

1
) = j

(< 0). Statement (2) can be equivalenty rewritten

as for all q [0, c]
(x

, j

) _
1
(q, x

1
, j

), for all q [0, c]

3
whereas statement (3) is simply
(x

, 0) _
1
(q, x

1
, 0), for all q [0, c] (4)
Let v(j) = (,j, ..., ,j, (1,)j) 1
N
, and use (1) to verify that (x

, 0) =
(x

, j

) + v(j

), (q, x

1
, 0) = (q, x

1
, j

) + v(j

), and apply supposition

PUT
1
to show that (x

, j

) _
1
(q, x

1
, j

) implies (x

, 0) _
1
(q, x

1
, 0).
Case (b): i(q, x

1
) j

. If i(q, x

1
) _ 0 then
(x

, i(q, x

1
)) _
1
(x

, j

)) _
1
(q, x

1
, i(q, x

1
)) (5)
where the rst weak preference statement follows from(x

, i(q, x

1
)) = (x

, j

)+
v(y

i(q, x

1
)) and supposition PUT whereas the second one follows from
statement (2). In this case c(q, x

1
) = 0 and verify that (x

, 0) = (q, x

1
, i(q, x

1
))+
v(i(q, x

1
)), (q, x

1
, 0) = (q, x

1
, i(q, x

1
)) +v(i(q, x

1
)), and apply sup-
position PUT
2
to show that statement 5 implies statement (3),
(x

, 0) _
1
(q, x

1
, 0) = (q, x

1
, c(q, x

1
)).
If i(q, x

1
) 0 then i(q, x

1
) = c(q, x

1
) and one has
(x

, 0) _
1
(x

, j

) _
1
(q, x

1
, i(q, x

1
)) = (q, x

1
, c(q, x

1
))
where the rst preference follows from (x

, 0) = (x

, j

) +v(j

) and suppo-
sition PUT whereas the second one follows from statement (2).
Thus, player 1 cannot do strictly better by deviating. Similarly for all other
A-players.
Proof of Proposition 4. Assume reciprocal preferences. Let g
N

i<N
G
i
,
be given. In the payo space, the opportunity sets of 1 -player at information
set J(g
N
) in the provision game and at information set J(e g
N
) in the
appropriation game are identical. We need to show that /r
P
(g
N
) e/r
A
(e
g
N
). Let q
N
be the most preferred choice of 1 player when information set
J(g
N
) is not chosen by A-players, that is q
N
is the optimal choice for xed
preferences. Let the corresponding vector of payos be = (
1
(g), ...,
N
(g)).
In the provision game, J(g
N
) 'GT J(0), because (a) the largest possible
payo for Y-player is larger at J(g
N
) than at J(0),
N
(g
N
) = c +,
P
i
q
i

c =
N
(0) and that (b) for any X-player , < ,
N
(g
j
, q
j
)
N
(g
j
, 0) =
,q
j
_
j
(g
j
, q
j
)
j
(g
j
, 0) = (, 1)q
j
. By Axioms o and 1, choice g
N
induces 'T preferences on 1 -player, which require that he (weakly) reduces
his optimal payo (according to unconditional preferences),
N
(g) to (weakly)
increase the payos,
i
(g) of X-player i. Y-player can do so by increasing his
contribution q
N
, hence,
/r
P
(g
N
) _ q
N
.
1
Denition (4) applies since y

0 and y

(N 1) + y

(1 ) 0:
2
Denition (4) applies since (g; x

1
) 0 and (g; x

1
)(N1) +(g; x

1
)(1) 0:
4
On the other hand, in the approriation game, J(e 0) 'GT J(e g
N
), and
Axioms o and 1 require that player 1 (weakly) decrease the payos of others,

N
(g); 1 -player can do so by increasing his appropriation
/r
A
(e g
N
) _ c q
N
.
The last two inequalities imply that /r
P
(g
N
) _ c /r
A
(e g
N
).
5